[aarch64-port-dev ] RFR (XS) 8252366: Shenandoah: revert/cleanup changes in graphKit.cpp

[Roman Kennke]

Am Mittwoch, den 26.08.2020, 13:08 +0200 schrieb Aleksey Shipilev:
> On 8/26/20 1:01 PM, Roland Westrelin wrote:
> > > It looks like this is not actually used for reference types. The
> > > only
> > > uses are in library_call.cpp and they don't look like they ever
> > > load a
> > > reference. Please check and confirm to be sure.
> > 
> > It does on PPC. In
> > LibraryCallKit::get_key_start_from_aescrypt_object().
> 
> Yes. I misjudged the assert: it actually verifies the current
> Shenandoah uses do not need the LRB 
> there. Shenandoah is not supported (yet) on PPC, so there is no bug
> yet, and assert is just too 
> strong for PPC code. But it feels more future proof just to insert
> the Shenandoah barrier there.
> 
> 8u webrev:
>   https://cr.openjdk.java.net/~shade/8252366/webrev.02/
> 
> Testing: hotspot_gc_shenandoah {fastdebug,release}

Looks good.

It's unfortunate that we currently have no way to verify if it works or
not, though. It's currently dead code and if we ever backport an
intrinsic that needs it, we hope for the best.

Roman